This paper contains supplementary material that can be found online at http://journals.cambridge.org
Introduction
The concept of large-scale habitat connectivity via protected areas and associated wildlife corridors has become a driving force in conservation of the jaguar Panthera onca (Sanderson et al., Reference Sanderson, Redford, Chetkiewicz, Medellin, Rabinowitz, Robinson and Taber2002; Rabinowitz & Zeller, Reference Rabinowitz and Zeller2010). As human population growth continues, wildlife corridors are considered a means to prevent the loss of genetic diversity in resident wildlife populations as a result of habitat fragmentation (Beier & Noss, Reference Beier and Noss1998). The jaguar has become a model species for the implementation of such strategies in Mesoamerica because of its large home range requirement, threatened status and significant role as a keystone species and apex predator (Swank & Teer, Reference Swank and Teer1989; Sunquist & Sunquist, Reference Sunquist and Sunquist2002). A recent analysis of mitochondrial DNA from blood samples across the species’ range showed high gene flow and little evidence of geographic barriers to dispersal (Eizirik et al., Reference Eizirik, Kim, Menotti-Raymond, Crawshaw, O'Brien and Johnson2001), suggesting that large-scale habitat connectivity may still exist.
The purpose of corridors is to promote species persistence via dispersal and subsequent genetic exchange (Noss, Reference Noss1987). As habitat fragmentation increases and gene flow is reduced or prevented, potential effects include a smaller effective population size and a reduction in adaptive fitness as a result of genetic drift and inbreeding (MacArthur & Wilson, Reference MacArthur and Wilson1967; Soule & Mills, Reference Soule and Mills1998). The effects of reduced genetic exchange in felids have been well documented, particularly in relation to the Florida panther Felis concolor coryi (Hedrick, Reference Hedrick1995). Isolated populations also suffer the negative effects of demographic and environmental stochasticity (Brown & Kodric-Brown, Reference Brown and Kodric-Brown1977).
Jaguars have been shown to inhabit human-impacted areas with varying degrees of disturbance (Foster et al., Reference Foster, Harmsen and Doncaster2010) but little is known about the status of jaguar populations outside protected areas (Sanderson et al., Reference Sanderson, Redford, Chetkiewicz, Medellin, Rabinowitz, Robinson and Taber2002). Mexico, which is at the northernmost limit of permanent jaguar range, is no exception, with most jaguar research there occurring only recently (Monroy-Vilchis et al., Reference Monroy-Vilchis, Sanchez, Aguilera-Reyes, Suarez and Urios2008, Reference Monroy-Vilchis, Rodriguez-Soto, Zarco-Gonzalez and Urios2009; Grigione et al., Reference Grigione, Menke, Lopez-Gonzalez, List, Banda and Carrera2009).
The range-wide analysis conducted by Rabinowitz & Zeller (Reference Rabinowitz and Zeller2010) was one of the first modelling exercises of jaguar habitat across its range, using six landscape characteristics to identify least-cost corridors connecting 90 known jaguar populations. As this product was derived only from geospatial analysis and expert assignment of cost values, Zeller et al. (Reference Zeller, Nijhawan, Salom-Pérez, Potosme and Hines2011) used a field-based interview method to validate jaguar and prey presence in one of the least-cost corridors in Nicaragua. The interview data were analysed using site occupancy modelling, which had a simple detection/non-detection data requirement and the ability to model occupancy as a function of environmental covariates (MacKenzie et al., Reference MacKenzie, Nichols, Royle, Pollock, Bailey and Hines2006). The protocol allowed rapid, large-scale monitoring of jaguars and their prey while compensating for imperfect detection, and allowed the authors to refine a geospatially-derived corridor using field data (MacKenzie et al., Reference MacKenzie, Nichols, Seamans and Gutiérrez2009; Zeller et al., Reference Zeller, Nijhawan, Salom-Pérez, Potosme and Hines2011).
Rabinowitz & Zeller (Reference Rabinowitz and Zeller2010) suggested habitat connectivity along Mexico's Pacific coast via a 1,000 km corridor connecting jaguar populations in Jalisco and the Isthmus of Tehuantepec (Fig. 1). The only major jaguar population of the east coast, in the Sierra Madre Oriental, is connected to populations in Sonora and Jalisco via a tenuous 800 km corridor. Rabinowitz & Zeller (Reference Rabinowitz and Zeller2010) did not show any functional connectivity between the Sierra Madre Oriental and the Isthmus of Tehuantepec, a distance of c. 700 km.
Fig. 1 Location of jaguar Panthera onca populations and modelled least-cost corridors, from Rabinowitz & Zeller (Reference Rabinowitz and Zeller2010). Our study assessed the suitability of a jaguar corridor along Mexico's east coast, where habitat connectivity is unknown.
A large-scale analysis driven solely by geospatial data and expert assignment of cost values may not adequately represent on-the-ground conditions and could therefore overlook viable corridors. In response to this perceived shortcoming, as well as published accounts of jaguar presence in east-central Mexico (Espino-Barrios et al., Reference Espino-Barrios, Viera, Martinez, Zepeda, Correo and Torres2005; Monroy-Vilchis et al., Reference Monroy-Vilchis, Sanchez, Aguilera-Reyes, Suarez and Urios2008, Reference Monroy-Vilchis, Rodriguez-Soto, Zarco-Gonzalez and Urios2009), we proposed a potential corridor connecting jaguar populations of the Sierra Madre Oriental and the Isthmus of Tehuantepec (Fig. 1). Our study area was located in the centre of this proposed corridor (Fig. 2). Our goals were to determine jaguar presence in a little-known area of Mexico and to ascertain whether Rabinowitz & Zeller (Reference Rabinowitz and Zeller2010) omitted a functional corridor. We hypothesized that if connectivity remained along the east coast of Mexico there would be strong evidence of a viable jaguar population or frequent dispersers in the area.
Fig. 2 The study area in north-east Puebla state, Mexico (Fig. 1).
Study area
Our study area (5,040 km2) was located in north-east Puebla state, in east-central Mexico (Fig. 2). The area is characterized by a north-east to south-west elevation gradient of 50–2,840 m. The primary land-cover types are vegetation/cropland mosaic (40%), cropland and cropland/vegetation mosaic (35%) and savannah (13%). Approximately one-third of the protected area Cuenca Hidrográfica del Río Necaxa (total area 396 km2) lies within the western border of the study area. There are c. 730,000 inhabitants across 1,800 settlements in the region (median population size of settlements = 107), mostly comprising coffee, pepper and citrus fruit-growers.
Methods
Sampling design
We divided the study area into 140 6 × 6 km sampling units. The sampling unit size (36 km2) was based on the minimum estimated jaguar home ranges in the region: 25 and 32 km2 in Jalisco and Campeche states, respectively (Ceballos et al., Reference Ceballos, Chavez, Rivera, Manterola, Wall, Medellin, Equihua, Chetkiewicz, Crawshaw, Rabinowitz and Redford2002; Nunez et al., Reference Nunez, Miller, Lindzey, Medellin, Equihua, Chetkiewicz, Crawshaw, Rabinowitz and Redford2002). We conducted interviews with local inhabitants over a 5-month period, using the protocol outlined by Zeller et al. (Reference Zeller, Nijhawan, Salom-Pérez, Potosme and Hines2011). Each interview from the same sampling unit was considered a separate replicate for the computation of detection probabilities.
Interview protocol
The questionnaire was designed to gather detection histories for the jaguar and five common prey species: collared peccary Pecari tajacu, red brocket deer Mazama americana, white-tailed deer Odocoileus virginianus, spotted paca Agouti paca, and nine-banded armadillo Dasypus novemcinctus (Weckel et al., Reference Weckel, Giuliano and Silver2006).
The interview protocol specifically targeted those who were knowledgeable about wildlife in the area, such as hunters, farmers and ranchers. Interviewees also had to have known the area for at least 1 year and have visited the area at least twice per month. The first task of the interviewer was to determine the interviewee's area of knowledge, which could span one or more sampling units. The interview questions were specific to a single sampling unit, such that the questions had to be repeated if the interviewee's area of knowledge extended to more than one sampling unit. This ensured a separate detection history for each unit.
We collected data on whether a jaguar had been positively detected in a given sampling unit within the previous year. Positive detections could include tracks, scat, vocalizations or direct sightings of a live or recently dead animal. Following a verbal description of what an interviewee had seen, we produced reference cards with photographs and tracks from various species and asked the interviewee to confirm what he or she had seen. If there was any doubt concerning the validity of the response, the data were discarded.
We also collected data on how frequently each of the five prey species was seen. Responses were categorized into four groups based on frequency of sighting: absent (not observed), rare (observed 1–5 times per year), moderately observed (> 5 times per year but < 2 times per month) and frequently observed (⩾ 2 times per month). If a species was absent or rarely observed we asked interviewees why the species was not seen more frequently.
Lastly, we collected personal data from the interviewees. These data included age, sex, reason for going to the sampling unit (hunting, farming, logging, etc.), means of transportation within the sampling unit, how long they had lived in the area and how often they visited the sampling unit. These variables were then used as covariates in our models. We also inquired about impending development projects and each interviewee's personal perception of jaguars, although these data were only used to supplement the analysis.
Covariates
Site occupancy modelling uses a detection/non-detection matrix for a given species to estimate two separate parameters: likelihood of habitat use and likelihood of species detection, the latter of which is related to an observer's ability to detect the animal (MacKenzie et al., Reference MacKenzie, Nichols, Royle, Pollock, Bailey and Hines2006). These parameters can also be modelled as a function of environmental covariates, variables that we hypothesized would influence a species' use of the landscape.
Site covariates, which influence the likelihood of habitat use by each target species, included the proportion of each sampling unit with a particular land cover (water, broadleaf forest, needleleaf forest, mixed forest, savannah, vegetation/crop mosaic, cropland, wetland), the proportion of total forest (broadleaf, needleleaf and mixed), the proportion of total water (water and wetland), mean elevation (m), maximum elevation (m), and distance (m) to river, road, village and protected area. Sampling covariates, which influence the likelihood of detection, included all site covariates, with the addition of interviewee-specific variables: effort (% of year in the field), hunter (yes/no), farmer (yes/no), and whether the interviewee travelled by foot within the sampling unit (yes/no). We could then use the model to investigate whether certain characteristics (e.g. spending more time in the field, status as a hunter, travelling by foot) increased species detection.
Land cover was derived from the GlobCover 300 m dataset (ESA, 2008). We obtained elevation data from the SRTM 90 m digital elevation database v. 4.0 (Jarvis et al., Reference Jarvis, Reuter, Nelson and Guevara2008) and protected area data from the World Database on Protected Areas (IUCN & UNEP, 2010). Layers for roads, rivers and settlements were derived from Mexico's portal for geographical information (CONABIO, 2011). All spatial data were compiled using ArcMap v. 10.0 (ESRI, Redlands, USA).
Data analysis
We used Presence v. 3.1 (Hines, Reference Hines2010) to analyse the data. Interview responses regarding the presence/absence of the jaguar and its five prey species were coded as detection/non-detection matrices. There was a maximum of 12 replicates per site, with one site lacking data.
Single-state and multi-state models
For a summary of the site occupancy modelling method, see Supplementary Material 1.
Naïve occupancy values, which equate to the proportion of sampling units in which a species was positively detected, were 0.04, 0.31, 0.52 and 0.29 for the jaguar, collared peccary, red brocket deer and white-tailed deer, respectively. We analysed data from these species in a single-state model.
Naïve occupancy of the paca was high (0.92), suggesting that this species was ubiquitous in the study area. To address high naïve occupancy, paca data were analysed using a multi-state model developed by Nichols et al. (Reference Nichols, Hines, MacKenzie, Seamans and Gutierrez2007), which converted the four frequency-of-detection categories (absent, rare, moderately observed and frequently observed) into abundance states (0, 1, 2 and 3, respectively). This is based on the assumption that the more frequently a species or its sign was observed, the higher its relative abundance in the sampling unit. We modelled Ψ-cond in state 3 (Ψ-condstate3), which is the likelihood of species presence in the highest abundance state, in our multi-state model.
Probably because of the high naïve occupancy of the armadillo (0.97), neither a single-state nor multi-state model could be fit to the data. This resulted in a single-state model with an altered detection history such that states 0, 1 and 2 were equal to 0 (absent) and state 3 was equal to 1 (present). Thus the armadillo was considered absent in states 0–2 and covariates were modelled only for state 3 within a single-state model.
Model selection and model averaging
We used Akaike's Information Criterion, using the small-sample correction (AICc), to rank the models. This method uses the principle of parsimony to produce a fitted model with the fewest necessary parameters (Burnham & Anderson, Reference Burnham and Anderson2002). We computed Akaike weights to determine the relative goodness of fit of each of the models. We eliminated models that contained statistically insignificant parameter estimates. In the event of multiple models with similar Akaike weights and significant and/or ecologically relevant covariates, we used model averaging to estimate probabilities of habitat use and detection (Burnham & Anderson, Reference Burnham and Anderson2002).
Predictive maps and corridor assessment
We divided prey species into two groups: Group I comprised the smaller species (spotted paca, armadillo) and Group II comprised the larger species (collared peccary, red brocket deer, white-tailed deer). We created two conditional statements to obtain the prey probability estimate, the probability of both Group I species and at least two Group II species using a given sampling unit. We multiplied Ψ-condGI, the probability that both smaller species use a given sampling unit (Supplementary Material 2, Equation 1), and Ψ-condGII, the probability that at least two of the larger prey species use that sampling unit (Supplementary Material 2, Equation 2).
We identified sampling units suitable for a jaguar corridor as those with a prey probability estimate of at least 0.90 (Zeller et al., Reference Zeller, Nijhawan, Salom-Pérez, Potosme and Hines2011).
Results
We conducted 245 interviews in the 140 sampling units during June–October 2010, with a mean of 5.5 interviews per site. Interviewees were aged 18–80 years old, with a mean age of 44 years. Ninety-six percent of interviewees were male. Interviewees spent a mean of 9.78 ± SD 7.73 days per month within the sampling units. Their main purposes in the study area were farming (37% of interviewees), cattle ranching (24%) and hunting (16%). Major impending development projects included road construction (52% of projects), oil drilling (19%) and mining (16%).
In response to why certain prey species were not seen more frequently, the most common reasons given were hunting by humans (57% of responses) and deforestation (24%).
Fifty-five percent of respondents had a negative attitude towards jaguars. The most common negative perceptions were fear (48% of responses) and a desire to kill jaguars (6%).
The most predictive model for the jaguar was the null model (Table 1), which predicts habitat use based solely on the detection/non-detection matrix and is used in the absence of significant site covariates. Likelihood of detection increased if the interviewee was a hunter.
Table 1 Most predictive models and untransformed coefficients of covariates for the jaguar Panthera onca and five prey species. Data for the jaguar, collared peccary, red brocket deer, and white-tailed deer were analysed using single-state models, spotted paca data were analysed using a multi-state model and armadillo data were analysed using a modified single-state model such that state 3 = presence and states 0–2 = absence.
1 Difference in AICc value relative to top model
2 MaxElev, maximum elevation in sampling unit; Hunter, whether or not the interviewee was a hunter (yes/no); DistPA, mean distance from sampling unit to edge of nearest protected area; Cropland, proportion of cropland in sampling unit; ByFoot, whether or not the interviewee explored the sampling unit by foot (yes/no); MeanElev, mean elevation of sampling unit; Needleleaf, proportion of needleleaf forest in sampling unit
3 Proportion of sampling units in which species was detected
4 For multistate models, coefficients are for state 3
Occupancy models for the five prey species were mostly driven by elevation (Table 1).
Using the null model for the jaguar and averaged top models for all five prey species we created maps of probability of habitat use by sampling unit (Fig. 3).
Fig. 3 Estimated probabilities of habitat use by jaguar and five prey species in the sampling units.
To identify the sampling units suitable for a jaguar corridor we used a minimum prey probability estimate of 0.90 (Zeller et al., Reference Zeller, Nijhawan, Salom-Pérez, Potosme and Hines2011). Of the 140 sampling units, 31 (22.14%) were suitable for a corridor (Fig. 4).
Fig. 4 (a) Probability of both the spotted paca and armadillo using habitat in each sampling unit; (b) probability of at least two of the three larger prey species using habitat in each sampling unit; and (c) probability of scenarios (a) and (b) occurring in each sampling unit.
Discussion
Species modelling revealed that some sampling units have an adequate prey base but lack jaguar sign. Jaguar sign of any type (direct sighting, track, scat, vocalization) was detected in only five of 140 (3.60%) sampling units during the previous year, which suggests that jaguars are rare in north-east Puebla state and probably do not have a resident population within the 5,040 km2 study area.
Elevation was a major driver in the detection of the large prey species and the habitat use of the small prey species. Our study area was characterized by a sharp elevation gradient, in which the lowlands were heavily cultivated and the highlands remained forested. For this reason detection of the collared peccary and red brocket deer increased with elevation, as both species prefer forested habitats (Altrichter & Boaglio, Reference Altrichter and Boaglio2004; Tejeda-Cruz et al., Reference Tejeda-Cruz, Naranjo, Cuaron, Perales and Cruz-Burguete2009). Detection of the collared peccary was also associated with proximity to protected areas, which emphasizes this species’ partiality to forest at a distance from human settlements (Altrichter & Boaglio, Reference Altrichter and Boaglio2004).
Contrary to results for the collared peccary and red brocket deer, detection of white-tailed deer decreased with elevation, as this species has been documented to use human-altered habitat more frequently than expected and avoid dense rainforest (Tejeda-Cruz et al., Reference Tejeda-Cruz, Naranjo, Cuaron, Perales and Cruz-Burguete2009). The fact that white-tailed deer habitat use was associated with increased croplands further supports this trend.
For the spotted paca there was a positive association between habitat use in the highest abundance state and greater maximum elevation, as the paca prefers forested habitat with access to fruits, nuts and seeds (Beletsky, Reference Beletsky2010). The association between increased detection and traversing the sampling unit by foot is obvious and suggests that walking an area will result in higher detection rates than biking or driving.
The armadillo had a positive association between habitat use and (1) lower mean elevation and (2) less needleleaf forest. Armadillos are habitat generalists capable of persisting in a variety of landscapes (Beletsky, Reference Beletsky2010), which could explain the observed trend of armadillos using disturbed habitat at lower elevations.
The lack of jaguar sign despite the apparent abundance of prey is a cause for concern. During a 12-month period only five positive jaguar detections (three tracks, two direct sightings) in an area of 5,040 km2 were reported in > 245 interviews. Given that carnivore home range is thought to increase with decreasing habitat quality (Logan & Sweanor, Reference Logan and Sweanor2001) and prey availability (Crawshaw & Quigley, Reference Crawshaw and Quigley1991), and that we based our sampling unit size on findings from more productive environments (Ceballos et al., Reference Ceballos, Chavez, Rivera, Manterola, Wall, Medellin, Equihua, Chetkiewicz, Crawshaw, Rabinowitz and Redford2002; Nunez et al., Reference Nunez, Miller, Lindzey, Medellin, Equihua, Chetkiewicz, Crawshaw, Rabinowitz and Redford2002), we probably underestimated the true home range size of jaguars in our study area. The use of sampling units smaller than the mean home range should have increased, rather than decreased, the detection probability in our area. The paucity of jaguar sightings by knowledgeable interviewees within such a broad area does not provide strong evidence of a resident jaguar population, nor does it suggest that this area is consistently traversed by dispersing individuals.
An additional concern is the possibility of false detection. Each jaguar detection was in a different sampling unit and therefore was not confirmed by other interviewees. One drawback of our method is that the highest observed state has no ambiguity; therefore if a jaguar has been detected at a site it is not possible for the site to be unoccupied (MacKenzie et al., Reference MacKenzie, Nichols, Seamans and Gutiérrez2009). This inability to account for false positives could suggest that jaguars are present when they are in fact absent from our study area. Another possibility is that these were true detections of a dispersing individual that did not have a permanent home range in the area. Thus, this region may not have a resident population but may be acting as a corridor for occasional dispersers.
If jaguars are present in the study area, large-scale development appears to be the biggest threat to the viability of the area as a jaguar corridor. Current and impending projects in the north (oil drilling, new road networks), centre (mining) and south (hydroelectric dam) will affect habitat connectivity for jaguars and other wildlife. Our prey habitat-use results, which identified isolated blocks of corridor surrounded by unsuitable units (Fig. 4), suggest fragmentation of available habitat. Of added significance is the predominantly negative attitude of the local people towards jaguars. In an area where people fear jaguars and some express a desire to kill them, educational outreach would be helpful to change their perspective and promote tolerance of jaguars. However, changed mindsets alone will not increase jaguar presence in this area.
Although there appears to be a prey base for jaguars in this region, the low frequency of jaguar sign, evidence of habitat fragmentation and impending large-scale development suggest that our study area does not hold great promise as a jaguar corridor. Conservation measures may be necessary to maintain and/or restore connectivity in this region.
Biographical sketches
Lisanne Petracca is a geospatial analyst who specializes in geographical information systems, remote sensing and site occupancy-based analyses of habitat use by the jaguar and its prey. O. Eric Ramírez-Bravo works on various carnivore conservation projects in central Mexico and is director of fauna-related projects for the NGO CREANATURA. Lorna Hernández-Santín is a wildlife biologist specializing in carnivore research and geographical information systems analysis. She also studies biodiversity within human-dominated landscapes, with a goal of reducing human influence on wildlife.
